Title: Ethernet adaptive link rate (ALR) : analysis of a MAC handshake protocol
Full Citation
Permanent Link: http://ufdc.ufl.edu/UF00094699/00001
 Material Information
Title: Ethernet adaptive link rate (ALR) : analysis of a MAC handshake protocol
Physical Description: Book
Language: English
Creator: Anand, Himanshu
Reardon, Casey
Subramaniyan, Rajagopal
George, Alan D.
Publisher: Anand et al.
Place of Publication: Gainesville, Fla.
Copyright Date: 2006
 Record Information
Bibliographic ID: UF00094699
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.


This item has the following downloads:

LCN2006a ( PDF )

Full Text

Ethernet Adaptive Link Rate (ALR):
Analysis of a MAC Handshake Protocol

Himanshu Anand, Casey Reardon, Rajagopal Subramaniyan, Alan D. George
High-performance Computing and Simulation (HCS) Research Laboratory
Department of Electrical and Computer Engineering, University of Florida
Gainesville, Florida 32611-6200
{anand, reardon, subraman, george}@hcs.ufl.edu


In this paper, a handshake protocol at the Medium
Access Control (A4C) layer is proposed and analyzed
for dynamically changing the link rate in the Network
Interface Card (NIC), ,u. hl.lj, to network utilization,
and thus decreasing average power consumption.
Simulation results show that this protocol can be used
to change link rate in Ethernet network devices without
causing user-perceivable delays.

1. Introduction

Edge devices, which are primarily comprised of
desktop computers, are the largest Internet-related
energy consumers. On an Ethernet link connecting a
PC to a first-level LAN switch, power consumption of
the link (NIC + switch) differs by approximately 4 W
between 1 Gbps and 100 Mbps link rates [1]. For 10
Gbps links, the power consumption is even higher.
This observation suggests that switching to lower link
rates during low utilization periods will result in
reducing the energy consumption in NICs. However,
packet delays can be a potential trade-off when NICs
operate at lower link rates. In this paper, we propose
and analyze a MAC handshake protocol for
dynamically switching link rates without causing any
user-perceivable delays. The protocol uses a dual-
threshold policy proposed in [2] for dictating when to
change link rates. Whenever the buffer occupancy
drops below the low threshold or rises above the high
threshold, the handshake mechanism is activated.

2. MAC Handshake Protocol

The protocol includes an exchange of two frames to
initiate a link rate change, viz. MAC control frame and
ACK/NACK frame. The frame format was adapted
from the existing PAUSE frame. The modifications
proposed are in the opcode field (to identify the frame
* This material is based upon work supported by the National
Science Foundation under Grant No. 0519951.

as a MAC control or ACK/NACK frame) and in the
parameter field (to indicate the link rate). While ALR
has been proposed for switching between more than
two link rates, we consider only two link rates here.
The procedure followed by the protocol to switch to
lower link rate can be described as follows. During the
link initialization, ALR capability is advertised through
Auto-Negotiation. If node A decides to change the link
rate, following the ALR policy, node A sends a MAC
control frame with the desired (low) rate in the
"parameter" field at the high link rate. Node A starts a
timer that expires if an ACK/NACK frame from node
B is not received within a specified time. If the timer
expires, node A sends the MAC control frame again.
Once node B receives the MAC control frame, it stops
sending further data frames, and checks whether it can
switch to the (low) link rate requested by node A
according to the ALR policy. If node B decides in the
affirmative, it sends an ACK frame to node A at the old
(high) rate. After sending the ACK, node B changes its
link rate to the desired value and resynchronizes its
clocks. After receiving the ACK, node A changes its
speed to the new (low) link rate, resynchronizes its
clocks, and resumes the data transmission at the new
link rate. If node B decides that it cannot switch the
link rate, it sends a NACK frame to node A at the old
link rate. If node A receives a NACK frame, it
resumes packet transmission at the old link rate again.
If node A decides to switch back to a higher rate, it
sends a MAC control frame and node B always replies
with an ACK in this case and resynchronizes its link.
The time required to resynchronize the clocks is
implementation-specific to the PHY transceiver chip.
For our simulation model, we fix the resynchronization
time as 512 PCI clock cycles (7.75 is for a clock with
66 MHz frequency), which is the time required to
synchronize the clock generators when switching to
active state (DO) from cold state (D3) in the case of
Intel Gigabit controllers.

3. Performance Evaluation of Protocol

Simulation models were developed using the
MLDesigner simulation tool to analyze the
performance of the MAC handshake protocol. We
modeled a system with a NIC and a switch linecard
both with ALR capability. Dummy packets are fed as
inputs into the models, whose sizes and inter-arrival
times are derived from traffic traces.
Traces were collected at the University of Florida
on a 1 Gbps link connecting a desktop PC and a switch
using the Ethereal packet capturing tool. Different
usage scenarios, such as typical Internet surfing
(UF_SRF: sparse and bursty traffic), typical Internet
surfing followed by a file transfer (UF_VAR, variable
traffic), large file transfer (UF_HDT: high data
transmission), and video streaming (UF_HDR: high
data reception to study the transmit FIFO state at the
switch line card from where the data is being
transmitted) were used for trace collection.
A synthetic traffic trace (UF_SYN) was also
generated by studying the characteristics of the
collected 1 Gbps traces. The aim of generating this
trace was to analyze the state of the output buffer and
formulate a technique to reduce the high number of
switches in cases of traffic that could induce frequent
link rate oscillations.
The size of the transmit FIFO queue is fixed at 64
KB as in most commercially available 1000Base-T
NICs. For each traffic trace, the mean packet delay is
calculated as the overall average of the difference
between the time at which a packet arrives at the NIC
and gets stored into the queue and the time at which the
packet reaches the switch. This delay value takes the
queuing delay, processing delay, and propagation delay
into account. The NIC includes a mechanism that
checks the buffer occupancy every 0.5 milliseconds.

4. Simulation Results

The mean packet delay values for all the traces are
shown in Table 1. The values of low threshold and
high threshold were fixed as 50% and 80% of the
buffer size, respectively. It is observed that the mean
packet delay is not user-perceivable and does not cause
any buffer overflow for any of the traces. For the
typical user trace UF_SRF, no link rate switches from
low to high rate were observed. Thus, the NIC could
operate in the low rate for almost 100% of the time of
its operation.
For the synthetic trace, we analyzed the effect of
increasing the gap between the low threshold value and
the high threshold value on the number of rate switches
and mean packet delay. The resulting values are
shown in Table 2. The idea is to make the NIC operate

in a single (high or low) link rate for a considerable
period of time once it has switched to that particular
rate. We observed that increasing the gap between the
high and low threshold values reduces the number of
switches without affecting the mean packet delay by a
considerable amount, while still leading to significant
power savings.

Table 1. Mean Packet Delay
Mean Packet Delay (ms) No. of low-
Trace to-high rate
1 Gbps 100 Mbps ALR switches
UF SRF 0.017 0.311 0.313 0
UF VAR 0.01 0.251 0.253 0
UF HDT 0.0325 Overflow 0.446 3
UF HDR 0.025 Overflow 0.901 10
UF SYN 0.0223 Overflow 0.832 46

Table 2. Threshold Analysis for UF SYN
High Low Mean No. of
threshold threshold Packet rate
(% of output (% of output Delay switches
buffer size) buffer size) (ms)
80 50 0.832 93
85 50 0.843 62
86 50 0.841 56
86 15 0.835 47
86 10 0.832 45

5. Conclusions

In this paper, we proposed and analyzed a MAC-
layer handshake mechanism to dynamically change the
link rate. The method is an effective mechanism for
changing link rates without causing any user-
perceivable packet delays. We also analyzed the MAC
handshake mechanism for traffic with periodically
changing link utilization with the help of a
synthetically generated trace. We observed a problem
of frequent link rate oscillations and proposed a
solution to this problem by increasing the gap between
high and low threshold values for the output buffer.


[1] C. Gunaratne, K. Christensen and B. Nordman,
"Managing energy consumption costs in desktop PCs
and LAN switches with proxying, split TCP
connections, and scaling of link speed," International
Journal of Network Management, Vol. 15, No. 5, pp.
297-310, September/October 2005.
[2] C. Gunaratne, K. Christensen, and S. Suen, "Ethernet
Adaptive Link Rate (ALR): Analysis of a buffer
threshold policy," to appear in IEEE GLOBECOM 2006.
liip '" cc I i c ..Ii /pgunarat/papers/globe06.pdf

University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs